Support conversion of bfp_b types directly to bfloat16

[jjiangTT]

There is no direct conversion from bfp_b types to the compact bfloat16 type and one is thus forced to go bfp_b->float->bfloat16 if they wish to use it.
The code seems very low level with simd instructions, shifting, masks, and other such extravagances so I can understand this being quite difficult though.